]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - drivers/kvm/svm.c
projects / mirror_ubuntu-zesty-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'master' of git://git.infradead.org/~dedekind/ubi-2.6
[mirror_ubuntu-zesty-kernel.git] / drivers / kvm / svm.c
1 /*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * AMD SVM support
5 *
6 * Copyright (C) 2006 Qumranet, Inc.
7 *
8 * Authors:
9 * Yaniv Kamay <yaniv@qumranet.com>
10 * Avi Kivity <avi@qumranet.com>
11 *
12 * This work is licensed under the terms of the GNU GPL, version 2. See
13 * the COPYING file in the top-level directory.
14 *
15 */
16
17 #include "kvm_svm.h"
18 #include "x86_emulate.h"
19 #include "irq.h"
20
21 #include <linux/module.h>
22 #include <linux/kernel.h>
23 #include <linux/vmalloc.h>
24 #include <linux/highmem.h>
25 #include <linux/sched.h>
26
27 #include <asm/desc.h>
28
29 MODULE_AUTHOR("Qumranet");
30 MODULE_LICENSE("GPL");
31
32 #define IOPM_ALLOC_ORDER 2
33 #define MSRPM_ALLOC_ORDER 1
34
35 #define DB_VECTOR 1
36 #define UD_VECTOR 6
37 #define GP_VECTOR 13
38
39 #define DR7_GD_MASK (1 << 13)
40 #define DR6_BD_MASK (1 << 13)
41
42 #define SEG_TYPE_LDT 2
43 #define SEG_TYPE_BUSY_TSS16 3
44
45 #define KVM_EFER_LMA (1 << 10)
46 #define KVM_EFER_LME (1 << 8)
47
48 #define SVM_FEATURE_NPT (1 << 0)
49 #define SVM_FEATURE_LBRV (1 << 1)
50 #define SVM_DEATURE_SVML (1 << 2)
51
52 static void kvm_reput_irq(struct vcpu_svm *svm);
53
54 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
55 {
56 return container_of(vcpu, struct vcpu_svm, vcpu);
57 }
58
59 unsigned long iopm_base;
60 unsigned long msrpm_base;
61
62 struct kvm_ldttss_desc {
63 u16 limit0;
64 u16 base0;
65 unsigned base1 : 8, type : 5, dpl : 2, p : 1;
66 unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
67 u32 base3;
68 u32 zero1;
69 } __attribute__((packed));
70
71 struct svm_cpu_data {
72 int cpu;
73
74 u64 asid_generation;
75 u32 max_asid;
76 u32 next_asid;
77 struct kvm_ldttss_desc *tss_desc;
78
79 struct page *save_area;
80 };
81
82 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
83 static uint32_t svm_features;
84
85 struct svm_init_data {
86 int cpu;
87 int r;
88 };
89
90 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
91
92 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
93 #define MSRS_RANGE_SIZE 2048
94 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
95
96 #define MAX_INST_SIZE 15
97
98 static inline u32 svm_has(u32 feat)
99 {
100 return svm_features & feat;
101 }
102
103 static inline u8 pop_irq(struct kvm_vcpu *vcpu)
104 {
105 int word_index = __ffs(vcpu->irq_summary);
106 int bit_index = __ffs(vcpu->irq_pending[word_index]);
107 int irq = word_index * BITS_PER_LONG + bit_index;
108
109 clear_bit(bit_index, &vcpu->irq_pending[word_index]);
110 if (!vcpu->irq_pending[word_index])
111 clear_bit(word_index, &vcpu->irq_summary);
112 return irq;
113 }
114
115 static inline void push_irq(struct kvm_vcpu *vcpu, u8 irq)
116 {
117 set_bit(irq, vcpu->irq_pending);
118 set_bit(irq / BITS_PER_LONG, &vcpu->irq_summary);
119 }
120
121 static inline void clgi(void)
122 {
123 asm volatile (SVM_CLGI);
124 }
125
126 static inline void stgi(void)
127 {
128 asm volatile (SVM_STGI);
129 }
130
131 static inline void invlpga(unsigned long addr, u32 asid)
132 {
133 asm volatile (SVM_INVLPGA :: "a"(addr), "c"(asid));
134 }
135
136 static inline unsigned long kvm_read_cr2(void)
137 {
138 unsigned long cr2;
139
140 asm volatile ("mov %%cr2, %0" : "=r" (cr2));
141 return cr2;
142 }
143
144 static inline void kvm_write_cr2(unsigned long val)
145 {
146 asm volatile ("mov %0, %%cr2" :: "r" (val));
147 }
148
149 static inline unsigned long read_dr6(void)
150 {
151 unsigned long dr6;
152
153 asm volatile ("mov %%dr6, %0" : "=r" (dr6));
154 return dr6;
155 }
156
157 static inline void write_dr6(unsigned long val)
158 {
159 asm volatile ("mov %0, %%dr6" :: "r" (val));
160 }
161
162 static inline unsigned long read_dr7(void)
163 {
164 unsigned long dr7;
165
166 asm volatile ("mov %%dr7, %0" : "=r" (dr7));
167 return dr7;
168 }
169
170 static inline void write_dr7(unsigned long val)
171 {
172 asm volatile ("mov %0, %%dr7" :: "r" (val));
173 }
174
175 static inline void force_new_asid(struct kvm_vcpu *vcpu)
176 {
177 to_svm(vcpu)->asid_generation--;
178 }
179
180 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
181 {
182 force_new_asid(vcpu);
183 }
184
185 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
186 {
187 if (!(efer & KVM_EFER_LMA))
188 efer &= ~KVM_EFER_LME;
189
190 to_svm(vcpu)->vmcb->save.efer = efer | MSR_EFER_SVME_MASK;
191 vcpu->shadow_efer = efer;
192 }
193
194 static void svm_inject_gp(struct kvm_vcpu *vcpu, unsigned error_code)
195 {
196 struct vcpu_svm *svm = to_svm(vcpu);
197
198 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
199 SVM_EVTINJ_VALID_ERR |
200 SVM_EVTINJ_TYPE_EXEPT |
201 GP_VECTOR;
202 svm->vmcb->control.event_inj_err = error_code;
203 }
204
205 static void inject_ud(struct kvm_vcpu *vcpu)
206 {
207 to_svm(vcpu)->vmcb->control.event_inj = SVM_EVTINJ_VALID |
208 SVM_EVTINJ_TYPE_EXEPT |
209 UD_VECTOR;
210 }
211
212 static int is_page_fault(uint32_t info)
213 {
214 info &= SVM_EVTINJ_VEC_MASK | SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
215 return info == (PF_VECTOR | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT);
216 }
217
218 static int is_external_interrupt(u32 info)
219 {
220 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
221 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
222 }
223
224 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
225 {
226 struct vcpu_svm *svm = to_svm(vcpu);
227
228 if (!svm->next_rip) {
229 printk(KERN_DEBUG "%s: NOP\n", __FUNCTION__);
230 return;
231 }
232 if (svm->next_rip - svm->vmcb->save.rip > MAX_INST_SIZE) {
233 printk(KERN_ERR "%s: ip 0x%llx next 0x%llx\n",
234 __FUNCTION__,
235 svm->vmcb->save.rip,
236 svm->next_rip);
237 }
238
239 vcpu->rip = svm->vmcb->save.rip = svm->next_rip;
240 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
241
242 vcpu->interrupt_window_open = 1;
243 }
244
245 static int has_svm(void)
246 {
247 uint32_t eax, ebx, ecx, edx;
248
249 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) {
250 printk(KERN_INFO "has_svm: not amd\n");
251 return 0;
252 }
253
254 cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
255 if (eax < SVM_CPUID_FUNC) {
256 printk(KERN_INFO "has_svm: can't execute cpuid_8000000a\n");
257 return 0;
258 }
259
260 cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
261 if (!(ecx & (1 << SVM_CPUID_FEATURE_SHIFT))) {
262 printk(KERN_DEBUG "has_svm: svm not available\n");
263 return 0;
264 }
265 return 1;
266 }
267
268 static void svm_hardware_disable(void *garbage)
269 {
270 struct svm_cpu_data *svm_data
271 = per_cpu(svm_data, raw_smp_processor_id());
272
273 if (svm_data) {
274 uint64_t efer;
275
276 wrmsrl(MSR_VM_HSAVE_PA, 0);
277 rdmsrl(MSR_EFER, efer);
278 wrmsrl(MSR_EFER, efer & ~MSR_EFER_SVME_MASK);
279 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
280 __free_page(svm_data->save_area);
281 kfree(svm_data);
282 }
283 }
284
285 static void svm_hardware_enable(void *garbage)
286 {
287
288 struct svm_cpu_data *svm_data;
289 uint64_t efer;
290 #ifdef CONFIG_X86_64
291 struct desc_ptr gdt_descr;
292 #else
293 struct Xgt_desc_struct gdt_descr;
294 #endif
295 struct desc_struct *gdt;
296 int me = raw_smp_processor_id();
297
298 if (!has_svm()) {
299 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
300 return;
301 }
302 svm_data = per_cpu(svm_data, me);
303
304 if (!svm_data) {
305 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
306 me);
307 return;
308 }
309
310 svm_data->asid_generation = 1;
311 svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
312 svm_data->next_asid = svm_data->max_asid + 1;
313 svm_features = cpuid_edx(SVM_CPUID_FUNC);
314
315 asm volatile ( "sgdt %0" : "=m"(gdt_descr) );
316 gdt = (struct desc_struct *)gdt_descr.address;
317 svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
318
319 rdmsrl(MSR_EFER, efer);
320 wrmsrl(MSR_EFER, efer | MSR_EFER_SVME_MASK);
321
322 wrmsrl(MSR_VM_HSAVE_PA,
323 page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
324 }
325
326 static int svm_cpu_init(int cpu)
327 {
328 struct svm_cpu_data *svm_data;
329 int r;
330
331 svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
332 if (!svm_data)
333 return -ENOMEM;
334 svm_data->cpu = cpu;
335 svm_data->save_area = alloc_page(GFP_KERNEL);
336 r = -ENOMEM;
337 if (!svm_data->save_area)
338 goto err_1;
339
340 per_cpu(svm_data, cpu) = svm_data;
341
342 return 0;
343
344 err_1:
345 kfree(svm_data);
346 return r;
347
348 }
349
350 static void set_msr_interception(u32 *msrpm, unsigned msr,
351 int read, int write)
352 {
353 int i;
354
355 for (i = 0; i < NUM_MSR_MAPS; i++) {
356 if (msr >= msrpm_ranges[i] &&
357 msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
358 u32 msr_offset = (i * MSRS_IN_RANGE + msr -
359 msrpm_ranges[i]) * 2;
360
361 u32 *base = msrpm + (msr_offset / 32);
362 u32 msr_shift = msr_offset % 32;
363 u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
364 *base = (*base & ~(0x3 << msr_shift)) |
365 (mask << msr_shift);
366 return;
367 }
368 }
369 BUG();
370 }
371
372 static __init int svm_hardware_setup(void)
373 {
374 int cpu;
375 struct page *iopm_pages;
376 struct page *msrpm_pages;
377 void *iopm_va, *msrpm_va;
378 int r;
379
380 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
381
382 if (!iopm_pages)
383 return -ENOMEM;
384
385 iopm_va = page_address(iopm_pages);
386 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
387 clear_bit(0x80, iopm_va); /* allow direct access to PC debug port */
388 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
389
390
391 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
392
393 r = -ENOMEM;
394 if (!msrpm_pages)
395 goto err_1;
396
397 msrpm_va = page_address(msrpm_pages);
398 memset(msrpm_va, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
399 msrpm_base = page_to_pfn(msrpm_pages) << PAGE_SHIFT;
400
401 #ifdef CONFIG_X86_64
402 set_msr_interception(msrpm_va, MSR_GS_BASE, 1, 1);
403 set_msr_interception(msrpm_va, MSR_FS_BASE, 1, 1);
404 set_msr_interception(msrpm_va, MSR_KERNEL_GS_BASE, 1, 1);
405 set_msr_interception(msrpm_va, MSR_LSTAR, 1, 1);
406 set_msr_interception(msrpm_va, MSR_CSTAR, 1, 1);
407 set_msr_interception(msrpm_va, MSR_SYSCALL_MASK, 1, 1);
408 #endif
409 set_msr_interception(msrpm_va, MSR_K6_STAR, 1, 1);
410 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_CS, 1, 1);
411 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_ESP, 1, 1);
412 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_EIP, 1, 1);
413
414 for_each_online_cpu(cpu) {
415 r = svm_cpu_init(cpu);
416 if (r)
417 goto err_2;
418 }
419 return 0;
420
421 err_2:
422 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
423 msrpm_base = 0;
424 err_1:
425 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
426 iopm_base = 0;
427 return r;
428 }
429
430 static __exit void svm_hardware_unsetup(void)
431 {
432 __free_pages(pfn_to_page(msrpm_base >> PAGE_SHIFT), MSRPM_ALLOC_ORDER);
433 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
434 iopm_base = msrpm_base = 0;
435 }
436
437 static void init_seg(struct vmcb_seg *seg)
438 {
439 seg->selector = 0;
440 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
441 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
442 seg->limit = 0xffff;
443 seg->base = 0;
444 }
445
446 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
447 {
448 seg->selector = 0;
449 seg->attrib = SVM_SELECTOR_P_MASK | type;
450 seg->limit = 0xffff;
451 seg->base = 0;
452 }
453
454 static void init_vmcb(struct vmcb *vmcb)
455 {
456 struct vmcb_control_area *control = &vmcb->control;
457 struct vmcb_save_area *save = &vmcb->save;
458
459 control->intercept_cr_read = INTERCEPT_CR0_MASK |
460 INTERCEPT_CR3_MASK |
461 INTERCEPT_CR4_MASK;
462
463 control->intercept_cr_write = INTERCEPT_CR0_MASK |
464 INTERCEPT_CR3_MASK |
465 INTERCEPT_CR4_MASK;
466
467 control->intercept_dr_read = INTERCEPT_DR0_MASK |
468 INTERCEPT_DR1_MASK |
469 INTERCEPT_DR2_MASK |
470 INTERCEPT_DR3_MASK;
471
472 control->intercept_dr_write = INTERCEPT_DR0_MASK |
473 INTERCEPT_DR1_MASK |
474 INTERCEPT_DR2_MASK |
475 INTERCEPT_DR3_MASK |
476 INTERCEPT_DR5_MASK |
477 INTERCEPT_DR7_MASK;
478
479 control->intercept_exceptions = 1 << PF_VECTOR;
480
481
482 control->intercept = (1ULL << INTERCEPT_INTR) |
483 (1ULL << INTERCEPT_NMI) |
484 (1ULL << INTERCEPT_SMI) |
485 /*
486 * selective cr0 intercept bug?
487 * 0: 0f 22 d8 mov %eax,%cr3
488 * 3: 0f 20 c0 mov %cr0,%eax
489 * 6: 0d 00 00 00 80 or $0x80000000,%eax
490 * b: 0f 22 c0 mov %eax,%cr0
491 * set cr3 ->interception
492 * get cr0 ->interception
493 * set cr0 -> no interception
494 */
495 /* (1ULL << INTERCEPT_SELECTIVE_CR0) | */
496 (1ULL << INTERCEPT_CPUID) |
497 (1ULL << INTERCEPT_HLT) |
498 (1ULL << INTERCEPT_INVLPGA) |
499 (1ULL << INTERCEPT_IOIO_PROT) |
500 (1ULL << INTERCEPT_MSR_PROT) |
501 (1ULL << INTERCEPT_TASK_SWITCH) |
502 (1ULL << INTERCEPT_SHUTDOWN) |
503 (1ULL << INTERCEPT_VMRUN) |
504 (1ULL << INTERCEPT_VMMCALL) |
505 (1ULL << INTERCEPT_VMLOAD) |
506 (1ULL << INTERCEPT_VMSAVE) |
507 (1ULL << INTERCEPT_STGI) |
508 (1ULL << INTERCEPT_CLGI) |
509 (1ULL << INTERCEPT_SKINIT) |
510 (1ULL << INTERCEPT_MONITOR) |
511 (1ULL << INTERCEPT_MWAIT);
512
513 control->iopm_base_pa = iopm_base;
514 control->msrpm_base_pa = msrpm_base;
515 control->tsc_offset = 0;
516 control->int_ctl = V_INTR_MASKING_MASK;
517
518 init_seg(&save->es);
519 init_seg(&save->ss);
520 init_seg(&save->ds);
521 init_seg(&save->fs);
522 init_seg(&save->gs);
523
524 save->cs.selector = 0xf000;
525 /* Executable/Readable Code Segment */
526 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
527 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
528 save->cs.limit = 0xffff;
529 /*
530 * cs.base should really be 0xffff0000, but vmx can't handle that, so
531 * be consistent with it.
532 *
533 * Replace when we have real mode working for vmx.
534 */
535 save->cs.base = 0xf0000;
536
537 save->gdtr.limit = 0xffff;
538 save->idtr.limit = 0xffff;
539
540 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
541 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
542
543 save->efer = MSR_EFER_SVME_MASK;
544
545 save->dr6 = 0xffff0ff0;
546 save->dr7 = 0x400;
547 save->rflags = 2;
548 save->rip = 0x0000fff0;
549
550 /*
551 * cr0 val on cpu init should be 0x60000010, we enable cpu
552 * cache by default. the orderly way is to enable cache in bios.
553 */
554 save->cr0 = 0x00000010 | X86_CR0_PG | X86_CR0_WP;
555 save->cr4 = X86_CR4_PAE;
556 /* rdx = ?? */
557 }
558
559 static void svm_vcpu_reset(struct kvm_vcpu *vcpu)
560 {
561 struct vcpu_svm *svm = to_svm(vcpu);
562
563 init_vmcb(svm->vmcb);
564 }
565
566 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
567 {
568 struct vcpu_svm *svm;
569 struct page *page;
570 int err;
571
572 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
573 if (!svm) {
574 err = -ENOMEM;
575 goto out;
576 }
577
578 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
579 if (err)
580 goto free_svm;
581
582 if (irqchip_in_kernel(kvm)) {
583 err = kvm_create_lapic(&svm->vcpu);
584 if (err < 0)
585 goto free_svm;
586 }
587
588 page = alloc_page(GFP_KERNEL);
589 if (!page) {
590 err = -ENOMEM;
591 goto uninit;
592 }
593
594 svm->vmcb = page_address(page);
595 clear_page(svm->vmcb);
596 svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
597 svm->asid_generation = 0;
598 memset(svm->db_regs, 0, sizeof(svm->db_regs));
599 init_vmcb(svm->vmcb);
600
601 fx_init(&svm->vcpu);
602 svm->vcpu.fpu_active = 1;
603 svm->vcpu.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
604 if (svm->vcpu.vcpu_id == 0)
605 svm->vcpu.apic_base |= MSR_IA32_APICBASE_BSP;
606
607 return &svm->vcpu;
608
609 uninit:
610 kvm_vcpu_uninit(&svm->vcpu);
611 free_svm:
612 kmem_cache_free(kvm_vcpu_cache, svm);
613 out:
614 return ERR_PTR(err);
615 }
616
617 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
618 {
619 struct vcpu_svm *svm = to_svm(vcpu);
620
621 __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
622 kvm_vcpu_uninit(vcpu);
623 kmem_cache_free(kvm_vcpu_cache, svm);
624 }
625
626 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
627 {
628 struct vcpu_svm *svm = to_svm(vcpu);
629 int i;
630
631 if (unlikely(cpu != vcpu->cpu)) {
632 u64 tsc_this, delta;
633
634 /*
635 * Make sure that the guest sees a monotonically
636 * increasing TSC.
637 */
638 rdtscll(tsc_this);
639 delta = vcpu->host_tsc - tsc_this;
640 svm->vmcb->control.tsc_offset += delta;
641 vcpu->cpu = cpu;
642 kvm_migrate_apic_timer(vcpu);
643 }
644
645 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
646 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
647 }
648
649 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
650 {
651 struct vcpu_svm *svm = to_svm(vcpu);
652 int i;
653
654 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
655 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
656
657 rdtscll(vcpu->host_tsc);
658 }
659
660 static void svm_vcpu_decache(struct kvm_vcpu *vcpu)
661 {
662 }
663
664 static void svm_cache_regs(struct kvm_vcpu *vcpu)
665 {
666 struct vcpu_svm *svm = to_svm(vcpu);
667
668 vcpu->regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
669 vcpu->regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
670 vcpu->rip = svm->vmcb->save.rip;
671 }
672
673 static void svm_decache_regs(struct kvm_vcpu *vcpu)
674 {
675 struct vcpu_svm *svm = to_svm(vcpu);
676 svm->vmcb->save.rax = vcpu->regs[VCPU_REGS_RAX];
677 svm->vmcb->save.rsp = vcpu->regs[VCPU_REGS_RSP];
678 svm->vmcb->save.rip = vcpu->rip;
679 }
680
681 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
682 {
683 return to_svm(vcpu)->vmcb->save.rflags;
684 }
685
686 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
687 {
688 to_svm(vcpu)->vmcb->save.rflags = rflags;
689 }
690
691 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
692 {
693 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
694
695 switch (seg) {
696 case VCPU_SREG_CS: return &save->cs;
697 case VCPU_SREG_DS: return &save->ds;
698 case VCPU_SREG_ES: return &save->es;
699 case VCPU_SREG_FS: return &save->fs;
700 case VCPU_SREG_GS: return &save->gs;
701 case VCPU_SREG_SS: return &save->ss;
702 case VCPU_SREG_TR: return &save->tr;
703 case VCPU_SREG_LDTR: return &save->ldtr;
704 }
705 BUG();
706 return NULL;
707 }
708
709 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
710 {
711 struct vmcb_seg *s = svm_seg(vcpu, seg);
712
713 return s->base;
714 }
715
716 static void svm_get_segment(struct kvm_vcpu *vcpu,
717 struct kvm_segment *var, int seg)
718 {
719 struct vmcb_seg *s = svm_seg(vcpu, seg);
720
721 var->base = s->base;
722 var->limit = s->limit;
723 var->selector = s->selector;
724 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
725 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
726 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
727 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
728 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
729 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
730 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
731 var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
732 var->unusable = !var->present;
733 }
734
735 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
736 {
737 struct vcpu_svm *svm = to_svm(vcpu);
738
739 dt->limit = svm->vmcb->save.idtr.limit;
740 dt->base = svm->vmcb->save.idtr.base;
741 }
742
743 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
744 {
745 struct vcpu_svm *svm = to_svm(vcpu);
746
747 svm->vmcb->save.idtr.limit = dt->limit;
748 svm->vmcb->save.idtr.base = dt->base ;
749 }
750
751 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
752 {
753 struct vcpu_svm *svm = to_svm(vcpu);
754
755 dt->limit = svm->vmcb->save.gdtr.limit;
756 dt->base = svm->vmcb->save.gdtr.base;
757 }
758
759 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
760 {
761 struct vcpu_svm *svm = to_svm(vcpu);
762
763 svm->vmcb->save.gdtr.limit = dt->limit;
764 svm->vmcb->save.gdtr.base = dt->base ;
765 }
766
767 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
768 {
769 }
770
771 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
772 {
773 struct vcpu_svm *svm = to_svm(vcpu);
774
775 #ifdef CONFIG_X86_64
776 if (vcpu->shadow_efer & KVM_EFER_LME) {
777 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
778 vcpu->shadow_efer |= KVM_EFER_LMA;
779 svm->vmcb->save.efer |= KVM_EFER_LMA | KVM_EFER_LME;
780 }
781
782 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG) ) {
783 vcpu->shadow_efer &= ~KVM_EFER_LMA;
784 svm->vmcb->save.efer &= ~(KVM_EFER_LMA | KVM_EFER_LME);
785 }
786 }
787 #endif
788 if ((vcpu->cr0 & X86_CR0_TS) && !(cr0 & X86_CR0_TS)) {
789 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
790 vcpu->fpu_active = 1;
791 }
792
793 vcpu->cr0 = cr0;
794 cr0 |= X86_CR0_PG | X86_CR0_WP;
795 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
796 svm->vmcb->save.cr0 = cr0;
797 }
798
799 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
800 {
801 vcpu->cr4 = cr4;
802 to_svm(vcpu)->vmcb->save.cr4 = cr4 | X86_CR4_PAE;
803 }
804
805 static void svm_set_segment(struct kvm_vcpu *vcpu,
806 struct kvm_segment *var, int seg)
807 {
808 struct vcpu_svm *svm = to_svm(vcpu);
809 struct vmcb_seg *s = svm_seg(vcpu, seg);
810
811 s->base = var->base;
812 s->limit = var->limit;
813 s->selector = var->selector;
814 if (var->unusable)
815 s->attrib = 0;
816 else {
817 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
818 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
819 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
820 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
821 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
822 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
823 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
824 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
825 }
826 if (seg == VCPU_SREG_CS)
827 svm->vmcb->save.cpl
828 = (svm->vmcb->save.cs.attrib
829 >> SVM_SELECTOR_DPL_SHIFT) & 3;
830
831 }
832
833 /* FIXME:
834
835 svm(vcpu)->vmcb->control.int_ctl &= ~V_TPR_MASK;
836 svm(vcpu)->vmcb->control.int_ctl |= (sregs->cr8 & V_TPR_MASK);
837
838 */
839
840 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
841 {
842 return -EOPNOTSUPP;
843 }
844
845 static int svm_get_irq(struct kvm_vcpu *vcpu)
846 {
847 struct vcpu_svm *svm = to_svm(vcpu);
848 u32 exit_int_info = svm->vmcb->control.exit_int_info;
849
850 if (is_external_interrupt(exit_int_info))
851 return exit_int_info & SVM_EVTINJ_VEC_MASK;
852 return -1;
853 }
854
855 static void load_host_msrs(struct kvm_vcpu *vcpu)
856 {
857 #ifdef CONFIG_X86_64
858 wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
859 #endif
860 }
861
862 static void save_host_msrs(struct kvm_vcpu *vcpu)
863 {
864 #ifdef CONFIG_X86_64
865 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
866 #endif
867 }
868
869 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *svm_data)
870 {
871 if (svm_data->next_asid > svm_data->max_asid) {
872 ++svm_data->asid_generation;
873 svm_data->next_asid = 1;
874 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
875 }
876
877 svm->vcpu.cpu = svm_data->cpu;
878 svm->asid_generation = svm_data->asid_generation;
879 svm->vmcb->control.asid = svm_data->next_asid++;
880 }
881
882 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
883 {
884 return to_svm(vcpu)->db_regs[dr];
885 }
886
887 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
888 int *exception)
889 {
890 struct vcpu_svm *svm = to_svm(vcpu);
891
892 *exception = 0;
893
894 if (svm->vmcb->save.dr7 & DR7_GD_MASK) {
895 svm->vmcb->save.dr7 &= ~DR7_GD_MASK;
896 svm->vmcb->save.dr6 |= DR6_BD_MASK;
897 *exception = DB_VECTOR;
898 return;
899 }
900
901 switch (dr) {
902 case 0 ... 3:
903 svm->db_regs[dr] = value;
904 return;
905 case 4 ... 5:
906 if (vcpu->cr4 & X86_CR4_DE) {
907 *exception = UD_VECTOR;
908 return;
909 }
910 case 7: {
911 if (value & ~((1ULL << 32) - 1)) {
912 *exception = GP_VECTOR;
913 return;
914 }
915 svm->vmcb->save.dr7 = value;
916 return;
917 }
918 default:
919 printk(KERN_DEBUG "%s: unexpected dr %u\n",
920 __FUNCTION__, dr);
921 *exception = UD_VECTOR;
922 return;
923 }
924 }
925
926 static int pf_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
927 {
928 u32 exit_int_info = svm->vmcb->control.exit_int_info;
929 struct kvm *kvm = svm->vcpu.kvm;
930 u64 fault_address;
931 u32 error_code;
932 enum emulation_result er;
933 int r;
934
935 if (!irqchip_in_kernel(kvm) &&
936 is_external_interrupt(exit_int_info))
937 push_irq(&svm->vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
938
939 mutex_lock(&kvm->lock);
940
941 fault_address = svm->vmcb->control.exit_info_2;
942 error_code = svm->vmcb->control.exit_info_1;
943 r = kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
944 if (r < 0) {
945 mutex_unlock(&kvm->lock);
946 return r;
947 }
948 if (!r) {
949 mutex_unlock(&kvm->lock);
950 return 1;
951 }
952 er = emulate_instruction(&svm->vcpu, kvm_run, fault_address,
953 error_code);
954 mutex_unlock(&kvm->lock);
955
956 switch (er) {
957 case EMULATE_DONE:
958 return 1;
959 case EMULATE_DO_MMIO:
960 ++svm->vcpu.stat.mmio_exits;
961 return 0;
962 case EMULATE_FAIL:
963 kvm_report_emulation_failure(&svm->vcpu, "pagetable");
964 break;
965 default:
966 BUG();
967 }
968
969 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
970 return 0;
971 }
972
973 static int nm_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
974 {
975 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
976 if (!(svm->vcpu.cr0 & X86_CR0_TS))
977 svm->vmcb->save.cr0 &= ~X86_CR0_TS;
978 svm->vcpu.fpu_active = 1;
979
980 return 1;
981 }
982
983 static int shutdown_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
984 {
985 /*
986 * VMCB is undefined after a SHUTDOWN intercept
987 * so reinitialize it.
988 */
989 clear_page(svm->vmcb);
990 init_vmcb(svm->vmcb);
991
992 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
993 return 0;
994 }
995
996 static int io_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
997 {
998 u32 io_info = svm->vmcb->control.exit_info_1; //address size bug?
999 int size, down, in, string, rep;
1000 unsigned port;
1001
1002 ++svm->vcpu.stat.io_exits;
1003
1004 svm->next_rip = svm->vmcb->control.exit_info_2;
1005
1006 string = (io_info & SVM_IOIO_STR_MASK) != 0;
1007
1008 if (string) {
1009 if (emulate_instruction(&svm->vcpu, kvm_run, 0, 0) == EMULATE_DO_MMIO)
1010 return 0;
1011 return 1;
1012 }
1013
1014 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1015 port = io_info >> 16;
1016 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1017 rep = (io_info & SVM_IOIO_REP_MASK) != 0;
1018 down = (svm->vmcb->save.rflags & X86_EFLAGS_DF) != 0;
1019
1020 return kvm_emulate_pio(&svm->vcpu, kvm_run, in, size, port);
1021 }
1022
1023 static int nop_on_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1024 {
1025 return 1;
1026 }
1027
1028 static int halt_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1029 {
1030 svm->next_rip = svm->vmcb->save.rip + 1;
1031 skip_emulated_instruction(&svm->vcpu);
1032 return kvm_emulate_halt(&svm->vcpu);
1033 }
1034
1035 static int vmmcall_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1036 {
1037 svm->next_rip = svm->vmcb->save.rip + 3;
1038 skip_emulated_instruction(&svm->vcpu);
1039 return kvm_hypercall(&svm->vcpu, kvm_run);
1040 }
1041
1042 static int invalid_op_interception(struct vcpu_svm *svm,
1043 struct kvm_run *kvm_run)
1044 {
1045 inject_ud(&svm->vcpu);
1046 return 1;
1047 }
1048
1049 static int task_switch_interception(struct vcpu_svm *svm,
1050 struct kvm_run *kvm_run)
1051 {
1052 pr_unimpl(&svm->vcpu, "%s: task switch is unsupported\n", __FUNCTION__);
1053 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1054 return 0;
1055 }
1056
1057 static int cpuid_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1058 {
1059 svm->next_rip = svm->vmcb->save.rip + 2;
1060 kvm_emulate_cpuid(&svm->vcpu);
1061 return 1;
1062 }
1063
1064 static int emulate_on_interception(struct vcpu_svm *svm,
1065 struct kvm_run *kvm_run)
1066 {
1067 if (emulate_instruction(&svm->vcpu, NULL, 0, 0) != EMULATE_DONE)
1068 pr_unimpl(&svm->vcpu, "%s: failed\n", __FUNCTION__);
1069 return 1;
1070 }
1071
1072 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1073 {
1074 struct vcpu_svm *svm = to_svm(vcpu);
1075
1076 switch (ecx) {
1077 case MSR_IA32_TIME_STAMP_COUNTER: {
1078 u64 tsc;
1079
1080 rdtscll(tsc);
1081 *data = svm->vmcb->control.tsc_offset + tsc;
1082 break;
1083 }
1084 case MSR_K6_STAR:
1085 *data = svm->vmcb->save.star;
1086 break;
1087 #ifdef CONFIG_X86_64
1088 case MSR_LSTAR:
1089 *data = svm->vmcb->save.lstar;
1090 break;
1091 case MSR_CSTAR:
1092 *data = svm->vmcb->save.cstar;
1093 break;
1094 case MSR_KERNEL_GS_BASE:
1095 *data = svm->vmcb->save.kernel_gs_base;
1096 break;
1097 case MSR_SYSCALL_MASK:
1098 *data = svm->vmcb->save.sfmask;
1099 break;
1100 #endif
1101 case MSR_IA32_SYSENTER_CS:
1102 *data = svm->vmcb->save.sysenter_cs;
1103 break;
1104 case MSR_IA32_SYSENTER_EIP:
1105 *data = svm->vmcb->save.sysenter_eip;
1106 break;
1107 case MSR_IA32_SYSENTER_ESP:
1108 *data = svm->vmcb->save.sysenter_esp;
1109 break;
1110 default:
1111 return kvm_get_msr_common(vcpu, ecx, data);
1112 }
1113 return 0;
1114 }
1115
1116 static int rdmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1117 {
1118 u32 ecx = svm->vcpu.regs[VCPU_REGS_RCX];
1119 u64 data;
1120
1121 if (svm_get_msr(&svm->vcpu, ecx, &data))
1122 svm_inject_gp(&svm->vcpu, 0);
1123 else {
1124 svm->vmcb->save.rax = data & 0xffffffff;
1125 svm->vcpu.regs[VCPU_REGS_RDX] = data >> 32;
1126 svm->next_rip = svm->vmcb->save.rip + 2;
1127 skip_emulated_instruction(&svm->vcpu);
1128 }
1129 return 1;
1130 }
1131
1132 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1133 {
1134 struct vcpu_svm *svm = to_svm(vcpu);
1135
1136 switch (ecx) {
1137 case MSR_IA32_TIME_STAMP_COUNTER: {
1138 u64 tsc;
1139
1140 rdtscll(tsc);
1141 svm->vmcb->control.tsc_offset = data - tsc;
1142 break;
1143 }
1144 case MSR_K6_STAR:
1145 svm->vmcb->save.star = data;
1146 break;
1147 #ifdef CONFIG_X86_64
1148 case MSR_LSTAR:
1149 svm->vmcb->save.lstar = data;
1150 break;
1151 case MSR_CSTAR:
1152 svm->vmcb->save.cstar = data;
1153 break;
1154 case MSR_KERNEL_GS_BASE:
1155 svm->vmcb->save.kernel_gs_base = data;
1156 break;
1157 case MSR_SYSCALL_MASK:
1158 svm->vmcb->save.sfmask = data;
1159 break;
1160 #endif
1161 case MSR_IA32_SYSENTER_CS:
1162 svm->vmcb->save.sysenter_cs = data;
1163 break;
1164 case MSR_IA32_SYSENTER_EIP:
1165 svm->vmcb->save.sysenter_eip = data;
1166 break;
1167 case MSR_IA32_SYSENTER_ESP:
1168 svm->vmcb->save.sysenter_esp = data;
1169 break;
1170 default:
1171 return kvm_set_msr_common(vcpu, ecx, data);
1172 }
1173 return 0;
1174 }
1175
1176 static int wrmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1177 {
1178 u32 ecx = svm->vcpu.regs[VCPU_REGS_RCX];
1179 u64 data = (svm->vmcb->save.rax & -1u)
1180 | ((u64)(svm->vcpu.regs[VCPU_REGS_RDX] & -1u) << 32);
1181 svm->next_rip = svm->vmcb->save.rip + 2;
1182 if (svm_set_msr(&svm->vcpu, ecx, data))
1183 svm_inject_gp(&svm->vcpu, 0);
1184 else
1185 skip_emulated_instruction(&svm->vcpu);
1186 return 1;
1187 }
1188
1189 static int msr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1190 {
1191 if (svm->vmcb->control.exit_info_1)
1192 return wrmsr_interception(svm, kvm_run);
1193 else
1194 return rdmsr_interception(svm, kvm_run);
1195 }
1196
1197 static int interrupt_window_interception(struct vcpu_svm *svm,
1198 struct kvm_run *kvm_run)
1199 {
1200 svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
1201 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
1202 /*
1203 * If the user space waits to inject interrupts, exit as soon as
1204 * possible
1205 */
1206 if (kvm_run->request_interrupt_window &&
1207 !svm->vcpu.irq_summary) {
1208 ++svm->vcpu.stat.irq_window_exits;
1209 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1210 return 0;
1211 }
1212
1213 return 1;
1214 }
1215
1216 static int (*svm_exit_handlers[])(struct vcpu_svm *svm,
1217 struct kvm_run *kvm_run) = {
1218 [SVM_EXIT_READ_CR0] = emulate_on_interception,
1219 [SVM_EXIT_READ_CR3] = emulate_on_interception,
1220 [SVM_EXIT_READ_CR4] = emulate_on_interception,
1221 /* for now: */
1222 [SVM_EXIT_WRITE_CR0] = emulate_on_interception,
1223 [SVM_EXIT_WRITE_CR3] = emulate_on_interception,
1224 [SVM_EXIT_WRITE_CR4] = emulate_on_interception,
1225 [SVM_EXIT_READ_DR0] = emulate_on_interception,
1226 [SVM_EXIT_READ_DR1] = emulate_on_interception,
1227 [SVM_EXIT_READ_DR2] = emulate_on_interception,
1228 [SVM_EXIT_READ_DR3] = emulate_on_interception,
1229 [SVM_EXIT_WRITE_DR0] = emulate_on_interception,
1230 [SVM_EXIT_WRITE_DR1] = emulate_on_interception,
1231 [SVM_EXIT_WRITE_DR2] = emulate_on_interception,
1232 [SVM_EXIT_WRITE_DR3] = emulate_on_interception,
1233 [SVM_EXIT_WRITE_DR5] = emulate_on_interception,
1234 [SVM_EXIT_WRITE_DR7] = emulate_on_interception,
1235 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
1236 [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
1237 [SVM_EXIT_INTR] = nop_on_interception,
1238 [SVM_EXIT_NMI] = nop_on_interception,
1239 [SVM_EXIT_SMI] = nop_on_interception,
1240 [SVM_EXIT_INIT] = nop_on_interception,
1241 [SVM_EXIT_VINTR] = interrupt_window_interception,
1242 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
1243 [SVM_EXIT_CPUID] = cpuid_interception,
1244 [SVM_EXIT_HLT] = halt_interception,
1245 [SVM_EXIT_INVLPG] = emulate_on_interception,
1246 [SVM_EXIT_INVLPGA] = invalid_op_interception,
1247 [SVM_EXIT_IOIO] = io_interception,
1248 [SVM_EXIT_MSR] = msr_interception,
1249 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
1250 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
1251 [SVM_EXIT_VMRUN] = invalid_op_interception,
1252 [SVM_EXIT_VMMCALL] = vmmcall_interception,
1253 [SVM_EXIT_VMLOAD] = invalid_op_interception,
1254 [SVM_EXIT_VMSAVE] = invalid_op_interception,
1255 [SVM_EXIT_STGI] = invalid_op_interception,
1256 [SVM_EXIT_CLGI] = invalid_op_interception,
1257 [SVM_EXIT_SKINIT] = invalid_op_interception,
1258 [SVM_EXIT_MONITOR] = invalid_op_interception,
1259 [SVM_EXIT_MWAIT] = invalid_op_interception,
1260 };
1261
1262
1263 static int handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1264 {
1265 struct vcpu_svm *svm = to_svm(vcpu);
1266 u32 exit_code = svm->vmcb->control.exit_code;
1267
1268 kvm_reput_irq(svm);
1269
1270 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
1271 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
1272 kvm_run->fail_entry.hardware_entry_failure_reason
1273 = svm->vmcb->control.exit_code;
1274 return 0;
1275 }
1276
1277 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
1278 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR)
1279 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
1280 "exit_code 0x%x\n",
1281 __FUNCTION__, svm->vmcb->control.exit_int_info,
1282 exit_code);
1283
1284 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
1285 || svm_exit_handlers[exit_code] == 0) {
1286 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1287 kvm_run->hw.hardware_exit_reason = exit_code;
1288 return 0;
1289 }
1290
1291 return svm_exit_handlers[exit_code](svm, kvm_run);
1292 }
1293
1294 static void reload_tss(struct kvm_vcpu *vcpu)
1295 {
1296 int cpu = raw_smp_processor_id();
1297
1298 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1299 svm_data->tss_desc->type = 9; //available 32/64-bit TSS
1300 load_TR_desc();
1301 }
1302
1303 static void pre_svm_run(struct vcpu_svm *svm)
1304 {
1305 int cpu = raw_smp_processor_id();
1306
1307 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1308
1309 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
1310 if (svm->vcpu.cpu != cpu ||
1311 svm->asid_generation != svm_data->asid_generation)
1312 new_asid(svm, svm_data);
1313 }
1314
1315
1316 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
1317 {
1318 struct vmcb_control_area *control;
1319
1320 control = &svm->vmcb->control;
1321 control->int_vector = irq;
1322 control->int_ctl &= ~V_INTR_PRIO_MASK;
1323 control->int_ctl |= V_IRQ_MASK |
1324 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1325 }
1326
1327 static void svm_set_irq(struct kvm_vcpu *vcpu, int irq)
1328 {
1329 struct vcpu_svm *svm = to_svm(vcpu);
1330
1331 svm_inject_irq(svm, irq);
1332 }
1333
1334 static void svm_intr_assist(struct kvm_vcpu *vcpu)
1335 {
1336 struct vcpu_svm *svm = to_svm(vcpu);
1337 struct vmcb *vmcb = svm->vmcb;
1338 int intr_vector = -1;
1339
1340 kvm_inject_pending_timer_irqs(vcpu);
1341 if ((vmcb->control.exit_int_info & SVM_EVTINJ_VALID) &&
1342 ((vmcb->control.exit_int_info & SVM_EVTINJ_TYPE_MASK) == 0)) {
1343 intr_vector = vmcb->control.exit_int_info &
1344 SVM_EVTINJ_VEC_MASK;
1345 vmcb->control.exit_int_info = 0;
1346 svm_inject_irq(svm, intr_vector);
1347 return;
1348 }
1349
1350 if (vmcb->control.int_ctl & V_IRQ_MASK)
1351 return;
1352
1353 if (!kvm_cpu_has_interrupt(vcpu))
1354 return;
1355
1356 if (!(vmcb->save.rflags & X86_EFLAGS_IF) ||
1357 (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) ||
1358 (vmcb->control.event_inj & SVM_EVTINJ_VALID)) {
1359 /* unable to deliver irq, set pending irq */
1360 vmcb->control.intercept |= (1ULL << INTERCEPT_VINTR);
1361 svm_inject_irq(svm, 0x0);
1362 return;
1363 }
1364 /* Okay, we can deliver the interrupt: grab it and update PIC state. */
1365 intr_vector = kvm_cpu_get_interrupt(vcpu);
1366 svm_inject_irq(svm, intr_vector);
1367 kvm_timer_intr_post(vcpu, intr_vector);
1368 }
1369
1370 static void kvm_reput_irq(struct vcpu_svm *svm)
1371 {
1372 struct vmcb_control_area *control = &svm->vmcb->control;
1373
1374 if ((control->int_ctl & V_IRQ_MASK)
1375 && !irqchip_in_kernel(svm->vcpu.kvm)) {
1376 control->int_ctl &= ~V_IRQ_MASK;
1377 push_irq(&svm->vcpu, control->int_vector);
1378 }
1379
1380 svm->vcpu.interrupt_window_open =
1381 !(control->int_state & SVM_INTERRUPT_SHADOW_MASK);
1382 }
1383
1384 static void svm_do_inject_vector(struct vcpu_svm *svm)
1385 {
1386 struct kvm_vcpu *vcpu = &svm->vcpu;
1387 int word_index = __ffs(vcpu->irq_summary);
1388 int bit_index = __ffs(vcpu->irq_pending[word_index]);
1389 int irq = word_index * BITS_PER_LONG + bit_index;
1390
1391 clear_bit(bit_index, &vcpu->irq_pending[word_index]);
1392 if (!vcpu->irq_pending[word_index])
1393 clear_bit(word_index, &vcpu->irq_summary);
1394 svm_inject_irq(svm, irq);
1395 }
1396
1397 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1398 struct kvm_run *kvm_run)
1399 {
1400 struct vcpu_svm *svm = to_svm(vcpu);
1401 struct vmcb_control_area *control = &svm->vmcb->control;
1402
1403 svm->vcpu.interrupt_window_open =
1404 (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
1405 (svm->vmcb->save.rflags & X86_EFLAGS_IF));
1406
1407 if (svm->vcpu.interrupt_window_open && svm->vcpu.irq_summary)
1408 /*
1409 * If interrupts enabled, and not blocked by sti or mov ss. Good.
1410 */
1411 svm_do_inject_vector(svm);
1412
1413 /*
1414 * Interrupts blocked. Wait for unblock.
1415 */
1416 if (!svm->vcpu.interrupt_window_open &&
1417 (svm->vcpu.irq_summary || kvm_run->request_interrupt_window)) {
1418 control->intercept |= 1ULL << INTERCEPT_VINTR;
1419 } else
1420 control->intercept &= ~(1ULL << INTERCEPT_VINTR);
1421 }
1422
1423 static void save_db_regs(unsigned long *db_regs)
1424 {
1425 asm volatile ("mov %%dr0, %0" : "=r"(db_regs[0]));
1426 asm volatile ("mov %%dr1, %0" : "=r"(db_regs[1]));
1427 asm volatile ("mov %%dr2, %0" : "=r"(db_regs[2]));
1428 asm volatile ("mov %%dr3, %0" : "=r"(db_regs[3]));
1429 }
1430
1431 static void load_db_regs(unsigned long *db_regs)
1432 {
1433 asm volatile ("mov %0, %%dr0" : : "r"(db_regs[0]));
1434 asm volatile ("mov %0, %%dr1" : : "r"(db_regs[1]));
1435 asm volatile ("mov %0, %%dr2" : : "r"(db_regs[2]));
1436 asm volatile ("mov %0, %%dr3" : : "r"(db_regs[3]));
1437 }
1438
1439 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
1440 {
1441 force_new_asid(vcpu);
1442 }
1443
1444 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
1445 {
1446 }
1447
1448 static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1449 {
1450 struct vcpu_svm *svm = to_svm(vcpu);
1451 u16 fs_selector;
1452 u16 gs_selector;
1453 u16 ldt_selector;
1454
1455 pre_svm_run(svm);
1456
1457 save_host_msrs(vcpu);
1458 fs_selector = read_fs();
1459 gs_selector = read_gs();
1460 ldt_selector = read_ldt();
1461 svm->host_cr2 = kvm_read_cr2();
1462 svm->host_dr6 = read_dr6();
1463 svm->host_dr7 = read_dr7();
1464 svm->vmcb->save.cr2 = vcpu->cr2;
1465
1466 if (svm->vmcb->save.dr7 & 0xff) {
1467 write_dr7(0);
1468 save_db_regs(svm->host_db_regs);
1469 load_db_regs(svm->db_regs);
1470 }
1471
1472 clgi();
1473
1474 local_irq_enable();
1475
1476 asm volatile (
1477 #ifdef CONFIG_X86_64
1478 "push %%rbx; push %%rcx; push %%rdx;"
1479 "push %%rsi; push %%rdi; push %%rbp;"
1480 "push %%r8; push %%r9; push %%r10; push %%r11;"
1481 "push %%r12; push %%r13; push %%r14; push %%r15;"
1482 #else
1483 "push %%ebx; push %%ecx; push %%edx;"
1484 "push %%esi; push %%edi; push %%ebp;"
1485 #endif
1486
1487 #ifdef CONFIG_X86_64
1488 "mov %c[rbx](%[svm]), %%rbx \n\t"
1489 "mov %c[rcx](%[svm]), %%rcx \n\t"
1490 "mov %c[rdx](%[svm]), %%rdx \n\t"
1491 "mov %c[rsi](%[svm]), %%rsi \n\t"
1492 "mov %c[rdi](%[svm]), %%rdi \n\t"
1493 "mov %c[rbp](%[svm]), %%rbp \n\t"
1494 "mov %c[r8](%[svm]), %%r8 \n\t"
1495 "mov %c[r9](%[svm]), %%r9 \n\t"
1496 "mov %c[r10](%[svm]), %%r10 \n\t"
1497 "mov %c[r11](%[svm]), %%r11 \n\t"
1498 "mov %c[r12](%[svm]), %%r12 \n\t"
1499 "mov %c[r13](%[svm]), %%r13 \n\t"
1500 "mov %c[r14](%[svm]), %%r14 \n\t"
1501 "mov %c[r15](%[svm]), %%r15 \n\t"
1502 #else
1503 "mov %c[rbx](%[svm]), %%ebx \n\t"
1504 "mov %c[rcx](%[svm]), %%ecx \n\t"
1505 "mov %c[rdx](%[svm]), %%edx \n\t"
1506 "mov %c[rsi](%[svm]), %%esi \n\t"
1507 "mov %c[rdi](%[svm]), %%edi \n\t"
1508 "mov %c[rbp](%[svm]), %%ebp \n\t"
1509 #endif
1510
1511 #ifdef CONFIG_X86_64
1512 /* Enter guest mode */
1513 "push %%rax \n\t"
1514 "mov %c[vmcb](%[svm]), %%rax \n\t"
1515 SVM_VMLOAD "\n\t"
1516 SVM_VMRUN "\n\t"
1517 SVM_VMSAVE "\n\t"
1518 "pop %%rax \n\t"
1519 #else
1520 /* Enter guest mode */
1521 "push %%eax \n\t"
1522 "mov %c[vmcb](%[svm]), %%eax \n\t"
1523 SVM_VMLOAD "\n\t"
1524 SVM_VMRUN "\n\t"
1525 SVM_VMSAVE "\n\t"
1526 "pop %%eax \n\t"
1527 #endif
1528
1529 /* Save guest registers, load host registers */
1530 #ifdef CONFIG_X86_64
1531 "mov %%rbx, %c[rbx](%[svm]) \n\t"
1532 "mov %%rcx, %c[rcx](%[svm]) \n\t"
1533 "mov %%rdx, %c[rdx](%[svm]) \n\t"
1534 "mov %%rsi, %c[rsi](%[svm]) \n\t"
1535 "mov %%rdi, %c[rdi](%[svm]) \n\t"
1536 "mov %%rbp, %c[rbp](%[svm]) \n\t"
1537 "mov %%r8, %c[r8](%[svm]) \n\t"
1538 "mov %%r9, %c[r9](%[svm]) \n\t"
1539 "mov %%r10, %c[r10](%[svm]) \n\t"
1540 "mov %%r11, %c[r11](%[svm]) \n\t"
1541 "mov %%r12, %c[r12](%[svm]) \n\t"
1542 "mov %%r13, %c[r13](%[svm]) \n\t"
1543 "mov %%r14, %c[r14](%[svm]) \n\t"
1544 "mov %%r15, %c[r15](%[svm]) \n\t"
1545
1546 "pop %%r15; pop %%r14; pop %%r13; pop %%r12;"
1547 "pop %%r11; pop %%r10; pop %%r9; pop %%r8;"
1548 "pop %%rbp; pop %%rdi; pop %%rsi;"
1549 "pop %%rdx; pop %%rcx; pop %%rbx; \n\t"
1550 #else
1551 "mov %%ebx, %c[rbx](%[svm]) \n\t"
1552 "mov %%ecx, %c[rcx](%[svm]) \n\t"
1553 "mov %%edx, %c[rdx](%[svm]) \n\t"
1554 "mov %%esi, %c[rsi](%[svm]) \n\t"
1555 "mov %%edi, %c[rdi](%[svm]) \n\t"
1556 "mov %%ebp, %c[rbp](%[svm]) \n\t"
1557
1558 "pop %%ebp; pop %%edi; pop %%esi;"
1559 "pop %%edx; pop %%ecx; pop %%ebx; \n\t"
1560 #endif
1561 :
1562 : [svm]"a"(svm),
1563 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
1564 [rbx]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RBX])),
1565 [rcx]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RCX])),
1566 [rdx]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RDX])),
1567 [rsi]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RSI])),
1568 [rdi]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RDI])),
1569 [rbp]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_RBP]))
1570 #ifdef CONFIG_X86_64
1571 ,[r8 ]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R8])),
1572 [r9 ]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R9 ])),
1573 [r10]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R10])),
1574 [r11]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R11])),
1575 [r12]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R12])),
1576 [r13]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R13])),
1577 [r14]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R14])),
1578 [r15]"i"(offsetof(struct vcpu_svm,vcpu.regs[VCPU_REGS_R15]))
1579 #endif
1580 : "cc", "memory" );
1581
1582 local_irq_disable();
1583
1584 stgi();
1585
1586 if ((svm->vmcb->save.dr7 & 0xff))
1587 load_db_regs(svm->host_db_regs);
1588
1589 vcpu->cr2 = svm->vmcb->save.cr2;
1590
1591 write_dr6(svm->host_dr6);
1592 write_dr7(svm->host_dr7);
1593 kvm_write_cr2(svm->host_cr2);
1594
1595 load_fs(fs_selector);
1596 load_gs(gs_selector);
1597 load_ldt(ldt_selector);
1598 load_host_msrs(vcpu);
1599
1600 reload_tss(vcpu);
1601
1602 svm->next_rip = 0;
1603 }
1604
1605 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
1606 {
1607 struct vcpu_svm *svm = to_svm(vcpu);
1608
1609 svm->vmcb->save.cr3 = root;
1610 force_new_asid(vcpu);
1611
1612 if (vcpu->fpu_active) {
1613 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
1614 svm->vmcb->save.cr0 |= X86_CR0_TS;
1615 vcpu->fpu_active = 0;
1616 }
1617 }
1618
1619 static void svm_inject_page_fault(struct kvm_vcpu *vcpu,
1620 unsigned long addr,
1621 uint32_t err_code)
1622 {
1623 struct vcpu_svm *svm = to_svm(vcpu);
1624 uint32_t exit_int_info = svm->vmcb->control.exit_int_info;
1625
1626 ++vcpu->stat.pf_guest;
1627
1628 if (is_page_fault(exit_int_info)) {
1629
1630 svm->vmcb->control.event_inj_err = 0;
1631 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
1632 SVM_EVTINJ_VALID_ERR |
1633 SVM_EVTINJ_TYPE_EXEPT |
1634 DF_VECTOR;
1635 return;
1636 }
1637 vcpu->cr2 = addr;
1638 svm->vmcb->save.cr2 = addr;
1639 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
1640 SVM_EVTINJ_VALID_ERR |
1641 SVM_EVTINJ_TYPE_EXEPT |
1642 PF_VECTOR;
1643 svm->vmcb->control.event_inj_err = err_code;
1644 }
1645
1646
1647 static int is_disabled(void)
1648 {
1649 u64 vm_cr;
1650
1651 rdmsrl(MSR_VM_CR, vm_cr);
1652 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
1653 return 1;
1654
1655 return 0;
1656 }
1657
1658 static void
1659 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
1660 {
1661 /*
1662 * Patch in the VMMCALL instruction:
1663 */
1664 hypercall[0] = 0x0f;
1665 hypercall[1] = 0x01;
1666 hypercall[2] = 0xd9;
1667 hypercall[3] = 0xc3;
1668 }
1669
1670 static void svm_check_processor_compat(void *rtn)
1671 {
1672 *(int *)rtn = 0;
1673 }
1674
1675 static struct kvm_x86_ops svm_x86_ops = {
1676 .cpu_has_kvm_support = has_svm,
1677 .disabled_by_bios = is_disabled,
1678 .hardware_setup = svm_hardware_setup,
1679 .hardware_unsetup = svm_hardware_unsetup,
1680 .check_processor_compatibility = svm_check_processor_compat,
1681 .hardware_enable = svm_hardware_enable,
1682 .hardware_disable = svm_hardware_disable,
1683
1684 .vcpu_create = svm_create_vcpu,
1685 .vcpu_free = svm_free_vcpu,
1686 .vcpu_reset = svm_vcpu_reset,
1687
1688 .prepare_guest_switch = svm_prepare_guest_switch,
1689 .vcpu_load = svm_vcpu_load,
1690 .vcpu_put = svm_vcpu_put,
1691 .vcpu_decache = svm_vcpu_decache,
1692
1693 .set_guest_debug = svm_guest_debug,
1694 .get_msr = svm_get_msr,
1695 .set_msr = svm_set_msr,
1696 .get_segment_base = svm_get_segment_base,
1697 .get_segment = svm_get_segment,
1698 .set_segment = svm_set_segment,
1699 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
1700 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
1701 .set_cr0 = svm_set_cr0,
1702 .set_cr3 = svm_set_cr3,
1703 .set_cr4 = svm_set_cr4,
1704 .set_efer = svm_set_efer,
1705 .get_idt = svm_get_idt,
1706 .set_idt = svm_set_idt,
1707 .get_gdt = svm_get_gdt,
1708 .set_gdt = svm_set_gdt,
1709 .get_dr = svm_get_dr,
1710 .set_dr = svm_set_dr,
1711 .cache_regs = svm_cache_regs,
1712 .decache_regs = svm_decache_regs,
1713 .get_rflags = svm_get_rflags,
1714 .set_rflags = svm_set_rflags,
1715
1716 .tlb_flush = svm_flush_tlb,
1717 .inject_page_fault = svm_inject_page_fault,
1718
1719 .inject_gp = svm_inject_gp,
1720
1721 .run = svm_vcpu_run,
1722 .handle_exit = handle_exit,
1723 .skip_emulated_instruction = skip_emulated_instruction,
1724 .patch_hypercall = svm_patch_hypercall,
1725 .get_irq = svm_get_irq,
1726 .set_irq = svm_set_irq,
1727 .inject_pending_irq = svm_intr_assist,
1728 .inject_pending_vectors = do_interrupt_requests,
1729 };
1730
1731 static int __init svm_init(void)
1732 {
1733 return kvm_init_x86(&svm_x86_ops, sizeof(struct vcpu_svm),
1734 THIS_MODULE);
1735 }
1736
1737 static void __exit svm_exit(void)
1738 {
1739 kvm_exit_x86();
1740 }
1741
1742 module_init(svm_init)
1743 module_exit(svm_exit)
ubuntu-zesty-kernel mirror